Autism Spectrum Disorder Prediction in Toddlers Using a K-Nearest Neighbor-Based Model

Abstract

Conventional diagnostic procedures for autism spectrum disorder (ASD) are often resource-intensive and may delay access to appropriate support. This study proposes a K-Nearest Neighbor (KNN)–based model to support early ASD screening in toddlers using questionnaire-derived behavioral data. The study utilized the Toddler Autism Screening Dataset (July 2018), consisting of 1,054 records of toddlers aged 12–36 months. The dataset includes Q-CHAT-10 behavioral screening items (A1–A10), an aggregated screening score (Qchat-10-Score), and selected demographic features. Data preprocessing involved categorical encoding and Min–Max normalization to ensure suitability for distance-based classification. The KNN model was developed using Euclidean distance and evaluated through standard classification metrics and confusion matrix analysis. The results indicate that the proposed model achieved an accuracy of 95.73%, a precision of 100%, a recall of 93.66%, and an F1-score of 96.72%. Feature correlation analysis further confirms the dominant role of behavioral indicators in ASD prediction. These findings suggest that a simple and interpretable KNN-based model can effectively function as an early screening decision support tool, assisting caregivers and practitioners in identifying toddlers who may require further clinical assessment